A0919
Title: A causal perspective on managing credit risk
Authors: Christopher Bockel-Rickermann - KU Leuven (Belgium) [presenting]
Abstract: For financial intermediaries, clients not repaying debt pose a substantial risk to their businesses. Therefore, credit risk scoring is a crucial task for lenders. To date, academic research has provided many approaches and methodologies to tackling this problem. However, much of the research has so far focused on calculating the default risk for a specific borrower-loan pair and not on the outcomes of potential counterfactuals. This is likely restricting the implementability of models in credit operations. Lenders typically have significant discretion about the terms of a credit before and during origination. Accurate methods and models to aid during origination, however, are scarce. Hence, the problem of loan origination and its impact on default risks is stated as a problem of causal inference. The ``debt ratio'' of a credit is considered a treatment and causal machine learning is used to identify the individualized effect of the debt ratio of clients on their default probability. The resulting model may help practitioners to balance client needs and business risks and is assessed on a semi-synthetic credit risk data set. The approach aims to provide a new perspective on credit risk research and new tools to both practitioners and researchers alike.
